Model-Based Machine Learning for Joint Digital Backpropagation and PMD Compensation

More generally, one may regard the entire communication system design as an end-to-end reconstruction task and jointly optimize transmitter and receiver NNs [1]. Both traditional [2-4] and end-to-end learning [5-7] have received considerable attention for optical fiber systems. However, the reliance on NNs as universal (but sometimes poorly understood) function approximators makes it difficult to incorporate existing domain knowledge or interpret the obtained solutions. Rather than relying on NNs, a different approach is to start from an existing model and parameterize it. For fiberoptic systems, this can be done for example by considering the split-step method (SSM) for numerically solving the nonlinear Schr odinger equation (NLSE).